MPEG Standards -> How MPEG-4 trade-offs affect design

But what are the trade-offs in the implementation of MPEG-4 video tools? And more specifically, how do they affect the designs of emerging handheld consumer systems such as wireless video viewers and digital camcorders?

The MPEG-4 video standard was primarily developed for low-bit-rate applications up to 384 kbits/second for CIF (352 x 288) resolution, which makes it well-suited for handheld digital video systems with simi lar bandwidth requirements.

Basically, MPEG-4 provides a large set of "tools" that can be selectively applied. The tools are divided into many overlapping sets, called profiles. The MPEG-4 Simple Profile (SP) provides commonly useful tools. It starts with H.263, the highly efficient video-compression standard developed for videophones.

Any MPEG-4 decoder can decode an H.263 baseline bit stream. MPEG-4 SP also provides other tools--for example, ac/dc prediction and error resilience.

Core and Main
Two other exemplary MPEG-4 profiles are Core Profile (CP) and Main Profile (MP). CP includes all the features of SP, supports CIF at bit rates up to 2 Mbits/s and provides additional tools for binary shape coding and P-VOP-based temporal scalability, among others. Similarly, MP subsumes CP, so that an MPEG-4 MP decoder handles a large assortment of MPEG-4 video and multimedia features, up to 38.4 Mbits/s and 1,920 x 1,088 resolution. Examples of MP tools include gray-scale alpha- shape coding and static sprites.

When developing an MPEG-4 video player, the designer selects which profile to use--SP, CP or MP--to meet the product requirements. In reviewing the handheld MPEG-4 systems evolving today, almost all are designed for MPEG-4 SP. The basic reason: Of the many MPEG-4 features, low-bit-rate video is most generally useful. Thus, we focus on the MPEG-4 SP tools.

When developing an MPEG-4 video encoding system such as a digital camcorder, designers have much more flexibility in selecting MPEG-4 features. This is because an MPEG-4 encoder can produce a valid bit stream without using all of the tools that are supported by the corresponding MPEG-4 SP decoder. While some handheld MPEG-4 encoder systems are using multiple MPEG-4 SP features, the basic H.263 encoder appears to be the most cost-effective choice. We estimate that using a good H.263 en coder provides about 90 percent of the value of a full MPEG-4 SP encoder, while reducing the complexity by about 25 percent.

Interestingly, MPEG-4 audio and system control aspects of the standard are generally not being adopted. Instead, proprietary multimedia format "wrappers" have become de facto requirements for Internet video applications, controlling access to the underlying MPEG-4 video format. This contrasts to the way that the JPEG image format has been adopted for open sharing of images--for example, among digital cameras, Web browsers, PCs and printers. However, the MPEG-4 video format is apparently evolving with controlled access via proprietary software.

Perhaps the open specification for streaming audio and video being developed by the Internet Streaming Media Alliance will focus products on a single format moving forward.

One-chip architectures
We see three one-chip architectural approaches being pursued for implementing MPEG-4 in low-power handheld systems.

For the fixed-function ASIC approach, circuits are dedicated to MPEG-4 encoding and/or decoding. The functionality is basically fixed, subject to register settings to configure parameters and I/O. A disadvantage of this approach is that the handheld product is limited to the particular MPEG-4 functions provided by the chip. For example, if MPEG-4 encoder requirements change after the ASIC has been manufactured, the product cannot be adapted.

The RISC approach uses a fixed-function coprocessor to make up for the insufficient digital signal processing performance of the RISC to achieve full MPEG-4 frame rate. Here, the RISC does as much of the job as it can, and the coprocessor is minimized to do just as much as needed. A limitation: The RISC has little room left for other functions.

In the DSP approach, the DSP is used with or without a programmable coprocesso r to implement MPEG-4 completely in software, while much of the DSP and the RISC are available to implement other functions. This approach transfers the complexity of MPEG-4 algorithms from circuits to software, but overcomes the flexibility limitations of using fixed-function MPEG-4 devices.

VHS quality ahead
While MPEG-4 is being deployed, the implications of other multimedia standards in development-MPEG-7, MPEG-21 and H.26L-for handheld embedded systems remain to be seen. Among these, H.26L is one that could change everything by offering a compelling new feature: about a 50 percent bit-rate reduction compared with MPEG-4 SP with the same quality.

Wireless communications bit rates are increasing up to about 384 kbits with third-generation technology, while the best-quality MPEG-4 CIF video requires about 1.5 Mbits/s (perceptually about the same as VHS quality). Thus the capability of H.26L methods to halve that bit rate to about 750 kbits/s suggests near-VHS-quality digital video-on-demand services for future wireless handheld video displays.

Apart from handhelds, MPEG-4 is also emerging in Internet server and broadband infrastructure, where real-time MPEG-2-to-MPEG-4 and MPEG-4-to-proprietary transcoders are useful to translate content between formats for video on demand.